Instance-level explanation

(fig:localDALEXsummary) Summary of three approaches to local model exploration and explanation.

(fig:localDALEXsummary) Summary of three approaches to local model exploration and explanation.

Introduction

Instance-level explainers help to understand how a model yields a prediction for a single observation. We can think about the following situations as examples:

A model is a function with a \(p\)-dimensional vector \(x\) as an argument. The plot of the value(s) of the function can be constructed in a \(p+1\)-dimensional space. An example with \(p=2\) is presented in Figure @ref(fig:cutsSurfaceReady). We will use it as an illustration of key ideas. The plot provides an information about the values of the function in the vicinity of point \(x^*\).

(fig:cutsSurfaceReady) Response surface for a model that is a function of two variables. We are interested in understanding the response of a model in a single point x*

(fig:cutsSurfaceReady) Response surface for a model that is a function of two variables. We are interested in understanding the response of a model in a single point x*

There are many different tools that may be used to explore the predictions of the model around a single point \(x^*\). In the following sections we will describe the most popular approaches. They can be divided into three classes.

(fig:cutsTechnikiReady) Illustration of different approaches to instance-level explanation. Panel A presents a What-If analysis with Ceteris Paribus profiles. The profiles plot the model response as a function of a value of a single variable, while keeping the values of all other explanatory variables fixed. Panel B illustrates the concept of local models. A simpler white-box model is fitted around the point of interest. It describes the local behaviour of the black-box model. Panel C illustrates the concept of variable attributions. Additive effects of each variable show how the model response differs from the average.

(fig:cutsTechnikiReady) Illustration of different approaches to instance-level explanation. Panel A presents a What-If analysis with Ceteris Paribus profiles. The profiles plot the model response as a function of a value of a single variable, while keeping the values of all other explanatory variables fixed. Panel B illustrates the concept of local models. A simpler white-box model is fitted around the point of interest. It describes the local behaviour of the black-box model. Panel C illustrates the concept of variable attributions. Additive effects of each variable show how the model response differs from the average.